Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

vAccSOL: Efficient and Transparent AI Vision Offloading for Mobile Robots

Adam Zahir, Michele Gucciardom Falk Selker, Anastasios Nanos, Kostis Papazafeiropoulos, Carlos J. Bernardos, Nicolas Weber, Roberto Gonzalez

Abstract

Mobile robots are increasingly deployed for inspection, patrol, and search-and-rescue operations, relying on computer vision for perception, navigation, and autonomous decision-making. However, executing modern vision workloads onboard is challenging due to limited compute resources and strict energy constraints. While some platforms include embedded accelerators, these are typically tied to proprietary software stacks, leaving user-defined workloads to run on resource-constrained companion computers. We present vAccSOL, a framework for efficient and transparent execution of AI-based vision workloads across heterogeneous robotic and edge platforms. vAccSOL integrates two components: SOL, a neural network compiler that generates optimized inference libraries with minimal runtime dependencies, and vAccel, a lightweight execution framework that transparently dispatches inference locally on the robot or to nearby edge infrastructure. This combination enables hardware-optimized inference and flexible execution placement without requiring modifications to robot applications. We evaluate vAccSOL on a real-world testbed with a commercial quadruped robot and twelve deep learning models covering image classification, video classification, and semantic segmentation. Compared to a PyTorch compiler baseline, SOL achieves comparable or better inference performance. With edge offloading, vAccSOL reduces robot-side power consumption by up to 80% and edge-side power by up to 60% compared to PyTorch, while increasing vision pipeline frame rate by up to 24x, extending the operating lifetime of battery-powered robots.

vAccSOL: Efficient and Transparent AI Vision Offloading for Mobile Robots

Abstract

Mobile robots are increasingly deployed for inspection, patrol, and search-and-rescue operations, relying on computer vision for perception, navigation, and autonomous decision-making. However, executing modern vision workloads onboard is challenging due to limited compute resources and strict energy constraints. While some platforms include embedded accelerators, these are typically tied to proprietary software stacks, leaving user-defined workloads to run on resource-constrained companion computers. We present vAccSOL, a framework for efficient and transparent execution of AI-based vision workloads across heterogeneous robotic and edge platforms. vAccSOL integrates two components: SOL, a neural network compiler that generates optimized inference libraries with minimal runtime dependencies, and vAccel, a lightweight execution framework that transparently dispatches inference locally on the robot or to nearby edge infrastructure. This combination enables hardware-optimized inference and flexible execution placement without requiring modifications to robot applications. We evaluate vAccSOL on a real-world testbed with a commercial quadruped robot and twelve deep learning models covering image classification, video classification, and semantic segmentation. Compared to a PyTorch compiler baseline, SOL achieves comparable or better inference performance. With edge offloading, vAccSOL reduces robot-side power consumption by up to 80% and edge-side power by up to 60% compared to PyTorch, while increasing vision pipeline frame rate by up to 24x, extending the operating lifetime of battery-powered robots.
Paper Structure (17 sections, 5 figures)

This paper contains 17 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Computation Offloading for Mobile Robots
  4. AI Model Acceleration for Edge Deployment
  5. Transparent Offloading Frameworks
  6. The vAccSOL Framework
  7. Overview
  8. Compile-Time Layer: Model Optimization with SOL
  9. Run-Time Layer: Transparent Execution with vAccel
  10. Experimental Evaluation
  11. Experimental Setup
  12. Robotic Vision Pipeline
  13. Evaluation Methodology
  14. Evaluation Results
  15. Frame Rate and Latency Analysis
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Landscape of optimization and offloading frameworks.
  • Figure 2: vAccSOL architecture
  • Figure 3: Experimental testbed.
  • Figure 4: Frame rate and total processing latency of the robot vision application for single-camera input. Top: achieved frame rate (higher values indicate better performance). Middle and bottom: latency breakdown under local and edge execution (lower is better). We compare Torch and SOL under local execution and edge offloading. Results are grouped by task: (a) image classification, (b) video action recognition, and (c) semantic segmentation.
  • Figure 5: Power consumption of the robot CPU, edge CPU, and edge GPU during local and remote inference. Annotations quantify robot-side power savings from edge offloading (top) and edge-side power savings of SOL compared to Torch (middle, bottom).